Generative Adversarial Networks in Histological Image Segmentation: A Systematic Literature Review

Abstract

Histological image analysis plays a crucial role in understanding and diagnosing various diseases, but manually segmenting these images is often complex, time-consuming, and heavily reliant on expert knowledge. Generative adversarial networks (GANs) have emerged as promising tools to assist in this task, enhancing the accuracy and efficiency of segmentation in histological images. This systematic literature review aims to explore how GANs have been utilized for segmentation in this field, highlighting the latest trends, key challenges, and opportunities for future research. The review was conducted across multiple digital libraries, including IEEE, Springer, Scopus, MDPI, and PubMed, with combinations of the keywords “generative adversarial network” or “GAN”, “segmentation” or “image segmentation” or “semantic segmentation”, and “histology” or “histological” or “histopathology” or “histopathological”. We reviewed 41 GAN-based histological image segmentation articles published between December 2014 and February 2025. We summarized and analyzed these papers based on the segmentation regions, datasets, GAN tasks, segmentation tasks, and commonly used metrics. Additionally, we discussed advantages, challenges, and future research directions. The analyzed studies demonstrated the versatility of GANs in handling challenges like stain variability, multi-task segmentation, and data scarcity—all crucial challenges in the analysis of histopathological images. Nevertheless, the field still faces important challenges, such as the need for standardized datasets, robust evaluation metrics, and better generalization across diverse tissues and conditions.